// the take
Nutter is a testing framework for Databricks notebooks, split into a runner library installed on the cluster and a CLI for triggering runs from a laptop or CI pipeline. It's aimed at data/ML engineers who want to write test_<notebook> fixtures against production-style notebooks and wire the results into Azure DevOps (or any CI that can shell out and parse JUnit XML).
Test discovery is dead simple: subclass NutterFixture and name methods before_/run_/assertion_/after_<testname>, no decorators or special registration needed. It ships both in-notebook (NutterFixtureParallelRunner) and CLI-level (--max_parallel_tests) parallel execution, plus pattern-matching and recursive test selection (nutter run dataload/src* --recursive) so suites scale without hand-maintained lists. The CLI exits non-zero on failure and can emit JUnit XML, so it drops cleanly into PublishTestResults@2 or any CI that understands JUnit.
Everything routes through dbutils.notebook.run/exit, so tests can't run outside a live Databricks cluster, and local debugging means manually commenting out result.exit(dbutils) and remembering to uncomment it before pushing - an easy way to accidentally break your pipeline. Test results travel back to the CLI as the notebook's exit value, which Databricks caps in size; a large suite with verbose assertion output can hit that ceiling. Parallel execution offers zero isolation - the docs explicitly warn you're responsible for avoiding race conditions when fixtures touch the same tables or files, the framework does nothing to prevent it. Adoption looks thin for a Microsoft-badged project (318 stars/50 forks), so check the issue backlog before committing a CI pipeline to it.
We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →